Customer Feedback
Part of DTC unit economics
Modelling repeat purchase without assuming it will happen
Build a repeat-purchase model from first-order cohorts, order contribution and a no-repeat case, with a hypothetical example.
Model repeat purchase by counting later-order contribution across all customers in an original first-purchase group, including those who never return. Compare it with a no-repeat case. It shows what repeat buying could add without assuming that a second order will rescue the first.
Define the group and observation window
Group customers by first-order date and follow each group for the same number of weeks or months. State how customer records are matched and check duplicates where practical. A second item in the first basket is not a second order.
Compare groups only at ages they have both reached. A group acquired three months ago has no observed twelve-month result.
Choose a window that suits the product’s purchase cycle and state what remains unseen. Apply the same rules for cancelled, refunded and replaced orders across groups.
Count contribution per original customer
For each order in the window, use retained revenue after discounts and refunds, less its product, payment, packing, delivery, return and order-specific support costs. Later orders may have a different basket or delivery charge. Add first and later order contributions, subtract the acquisition spend for the original group, then divide by the number of customers in that group.
Consider a hypothetical group of 100 first-time customers. Each first order contributes $35 before acquisition. Twenty customers each place one second order contributing $25; five of those twenty each place one third order contributing $20.
Total contribution before acquisition is 100 × $35 + 20 × $25 + 5 × $20 = $4,100, or $41 per original customer. With no repeat orders, it would be $35. The invented repeat orders add $6 per original customer, not $45 for every customer. No real cohort or repeat rate was measured.
Subtract the group’s acquisition spend separately. If customers arrived through different routes, keep those routes identifiable rather than applying one assumed acquisition cost to everyone.
Compare an honest zero with bounded cases
| Case | Input | Meaning |
|---|---|---|
| No repeat | First-order contribution only | Exposure if no later order arrives within the window. |
| Bounded repeat | Later orders observed for comparable customers | A scenario with its group, period and differences stated. |
| Higher repeat | A specified improvement assumption | A sensitivity check, not an achieved result. |
Do not fill a young group’s future months with the best older group’s results without considering changes to product, price, availability and acquisition route. A few early enthusiasts may not represent a broader audience.
A platform’s cohort report may show orders or spending, but those figures are not automatically contribution after order costs and acquisition.
Use the result for a spending decision
Compare first-order contribution after acquisition with the bounded contribution over the stated window. If the first order loses money, calculate the later-order contribution needed to recover the shortfall, and by when.
Compare that requirement with what comparable groups have actually done. If it exceeds the evidence, lower acquisition spend, improve order economics or acknowledge the risk explicitly.
Forecast orders are not earned revenue. Review the decision when the group reaches the stated age, and keep a cash forecast for advertising, stock and operating payments due in the meantime.
